TIME: Benchmarking Framework for Time-Series Foundation Models

The TIME benchmark provides a standardized evaluation suite for time-series foundation models and forecasting algorithms. Created by zqiao11 and updated in 2026, the repository facilitates performance comparisons across diverse temporal datasets and is integrated with Hugging Face.

Use Cases

• Evaluating zero-shot performance of time-series foundation models using standardized forecasting metrics
• Benchmarking time-series forecasting algorithms against diverse temporal patterns
• Comparing model efficiency across multiple time-series datasets within a unified framework

Strengths

• Specifically designed for time-series foundation model benchmarking
• Integrated with Hugging Face datasets for standardized data loading
• Focuses on modern forecasting evaluation metrics

Limitations

• Metadata does not specify the exact number of rows or features
• Primarily serves as a codebase for benchmarking rather than a single static file

Provenance

Source

zqiao11 (GitHub)

Freshness

Updated as of March 2026.